Bug#922821: [bumblebee] bumblebee "cannot enable discrete graphics card"

[Luca Boccassi]

On Wed, 2019-02-20 at 18:22 +0100, Giovanni Mascellani wrote:
> Package: bumblebee
> Version: 3.2.1-20
> Severity: normal
> 
> Hi,
> 
> (I am not sure bumblebee is the right package, but it is not clear to
> me
> whether the problem stays with bumblebee, nvidia drivers or what
> else;
> feel free to reassign)
> 
> my discrete graphics card is (according to lspci) a "NVIDIA
> Corporation
> GP107GLM [Quadro P2000 Mobile] [10de:1cba] (rev ff)", mounted on a
> Dell
> Precision 5530 laptop. If I configure bumblebee to use "bbswitch" as
> PM
> (which, I believe, is what I should do), optirun is unable to work:
> 
> $ optirun glxinfo
> [  383.435979] [ERROR]Cannot access secondary GPU - error: Could not
> enable discrete graphics card
> 
> [  383.435997] [ERROR]Aborting because fallback start is disabled.
> 
> After this command, module bbswitch is loaded and module nvidia is
> not.
> Also, these lines are sent to syslog:
> 
> Feb 20 18:11:12 amalgama bumblebeed[1044]: [  436.831559]
> [ERROR]Could
> not enable discrete graphics card
> Feb 20 18:11:12 amalgama kernel: [  436.860809] bbswitch: enabling
> discrete graphics
> Feb 20 18:11:12 amalgama kernel: [  436.860828] pci 0000:01:00.0:
> Refused to change power state, currently in D3
> 
> If I try manually to modprobe nvidia:
> 
> # modprobe nvidia
> modprobe: ERROR: could not insert 'nvidia_current': No such device
> modprobe: ERROR: ../libkmod/libkmod-module.c:979 command_do() Error
> running install command for nvidia
> modprobe: ERROR: could not insert 'nvidia': Operation not permitted
> 
> Also, the following lines are sent to syslog:
> 
> Feb 20 18:11:44 amalgama kernel: [  468.637860] version 39.2
> Feb 20 18:11:44 amalgama kernel: [  468.639277] ipmi device interface
> Feb 20 18:11:44 amalgama kernel: [  468.648299] nvidia: module
> license
> 'NVIDIA' taints kernel.
> Feb 20 18:11:44 amalgama kernel: [  468.648301] Disabling lock
> debugging
> due to kernel taint
> Feb 20 18:11:44 amalgama kernel: [  468.660738] nvidia-nvlink: Nvlink
> Core is being initialized, major device number 240
> Feb 20 18:11:44 amalgama kernel: [  468.661126] NVRM: This is a 64-
> bit
> BAR mapped above 4GB by the system
> Feb 20 18:11:44 amalgama kernel: [  468.661126] NVRM: BIOS or the
> Linux
> kernel, but the PCI bridge
> Feb 20 18:11:44 amalgama kernel: [  468.661126] NVRM: immediately
> upstream of this GPU does not define
> Feb 20 18:11:44 amalgama kernel: [  468.661126] NVRM: a matching
> prefetchable memory window.
> Feb 20 18:11:44 amalgama kernel: [  468.661128] NVRM: This may be due
> to
> a known Linux kernel bug.  Please
> Feb 20 18:11:44 amalgama kernel: [  468.661128] NVRM: see the README
> section on 64-bit BARs for additional
> Feb 20 18:11:44 amalgama kernel: [  468.661128] NVRM: information.
> Feb 20 18:11:44 amalgama kernel: [  468.661134] nvidia: probe of
> 0000:01:00.0 failed with error -1
> Feb 20 18:11:44 amalgama kernel: [  468.661156] NVRM: The NVIDIA
> probe
> routine failed for 1 device(s).
> Feb 20 18:11:44 amalgama kernel: [  468.661157] NVRM: None of the
> NVIDIA
> graphics adapters were initialized!
> Feb 20 18:11:44 amalgama kernel: [  468.727130] nvidia-nvlink:
> Unregistered the Nvlink Core, major device number 240
> 
> The only information I manage to find about this README is here[1],
> and
> mentions that I should upgrade to a new kernel. But I am already
> running
> basically the latest one (4.19.0-3-amd64), so this is probably a
> false
> trace.
> 
>  [1]
> https://us.download.nvidia.com/XFree86/Linux-x86_64/410.93/README/knownissues.html
> 
> However, I have noticed that if I disable PM in bumblebee (set
> PMMethod
> to none) and restart my system, the module bbswitch does not get
> loaded
> (correct), the module nvidia gets automatically loaded (probably
> correct), GDM3 starts with Xorg instead of Wayland (not really
> correct,
> but probably not fundamental) and optirun works, running perfectly
> FlightGear at top FPS. Optirun stops working after suspending the
> system, but seems to survive hibernation. Of course, if I re-enable
> bbswitch either within bumblebee configuration or manually loading
> the
> module, everything stops working until the next cold reboot.
> 
> I am running from an up-to-date unstable, in particular with nvidia
> drivers version 410.93-2. I am available for doing further tests, as
> I
> really would like to fix this problem and use my discrete card
> without
> having to cold-reboot.
> 
> Thanks for maintaining bumblebee!
> 
> Giovanni.

Some newer models, depending on the manufactorer and the motherboard,
do not work with bbswitch but work with the kernel's runtime power
management - which is better. It looks like to be the case for you.

So try to keep pmmethod to none, and to change the following to true:

AlwaysUnloadKernelDriver=false

It's a new option that was added recently - basically the kernel PM
should automatically turn the card on/off when the modules are
loaded/unloaded.

-- 
Kind regards,
Luca Boccassi
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 488 bytes
Desc: This is a digitally signed message part
URL: <http://alioth-lists.debian.net/pipermail/pkg-nvidia-devel/attachments/20190221/a25b1dfc/attachment-0001.sig>